mcp-security

To secure MCP tool handlers, constrain string parameters with maxLength or pattern validation, canonicalize file paths to a trusted base directory, and execute shell commands via argument lists instead of shell=True.

Preventing malicious inputs in MCP server definitions requires strict schema validation using enum constraints for string parameters, ensuring tool schemas reject unrestricted or unvalidated user data before it reaches handlers.

Managing secrets in MCP tool configurations safely means sourcing sensitive values from the environment or a dedicated secrets manager rather than embedding them directly in server definitions or handler code.

Yes, you can add audit logging to MCP tool invocations by configuring server definitions to record every handler call, ensuring traceability and accountability for tool access involving file system, shell, or network operations.

Yes, this approach specifically targets MCP servers exposing file system, shell, or network access in Claude Code by confining file paths to trusted directories and validating inputs to prevent exploitation of those handlers.

Unrestricted file access often occurs when tool schemas lack path canonicalization; restricting it requires confining file paths to a trusted base directory and applying strict input validation to prevent path traversal attacks.